JS: add isRelevant(succ) to flowStep predicate

[erik-krogh]

While evaluating #3019 I found a weird performance regression in the sqlteaching benchmark that was most notable with the js/insecure-randomness query.
It ran about 3 times slower than on master.

Turns out that all I needed to trigger the performance regression was to add 3 edges to the exploratory flow (the exploratoryFlowStep predicate), corresponding to 3 calls to array.filter().
Adding those 3 edges caused the exploratory flow to find ~200.000 data-flow nodes, instead of 44, which caused the query to run 3 times slower.
I only needed to add the edges in the exploratory flow to trigger the regression, and nowhere else.

But most of the extra time was not spend in the exploratory flow, so it seemed there was room for improvement.

With this one line change most, but not all, of the performance was regained.
(Performance went from about 245 seconds to about 95 seconds)

And the change seems to improve performance across all benchmarks (big evaluation pending).

Here are two small evaluations on some benchmarks:
https://git.semmle.com/erik/dist-compare-reports/tree/profiling-js-max.northeurope.cloudapp.azure.com_1584361183161
https://git.semmle.com/erik/dist-compare-reports/tree/profiling-js-asger.northeurope.cloudapp.azure.com_1584356931002

[asgerf]

Cool! I think the convention was that each kind of step checked isRelevant on its own, but I can see this has slipped in a few places.

[erik-krogh]

Evaluation looks very good.

[asgerf]

💪